The interaction of core modules as a basis for elucidating network behavior determining Parkinson's disease pathogenesis

In this perspective paper, we show how central aspects of the emergent systems behavior (and malfunctioning) in networks can be understood by systematically studying and examining interactions of key subnetworks as a starting point, demonstrating this in the instance of Parkinson's disease. In so doing, we highlight our systems perspective on analyzing important aspects of network behavior by considering key subnetworks and their interactions. We used focused systems analysis to analyze the interaction of two key subnetworks each containing alpha-synuclein (a core component of the Parkinson's disease network), and reveal the underlying systems landscape of the emergent behavior. This provides non-trivial insights into understanding the origins and key drivers of systems behavior, different ways of targeting nodes for treatment purposes, a basis for stratifying patient populations, and an illuminating platform for more detailed modeling, which it can be used in conjunction with. We also demonstrate how the basic framework can be built upon to examine the effect of dopamine compartmentalization. This approach represents a distinct way of dissecting nonlinear networks and can be adapted and used in other disease contexts as well. Parkinson's disease is multifactorial, with variability in its pathogenesis/progression and no cure yet.1-4 This motivates the need to carefully dissect the underlying network, which involves a core component alpha-synuclein participating in two interacting subnetworks, each characterized by nonlinearity/feedback.5, 6 We present a framework (applicable in different disease contexts) for elucidating emergent behavior from the interaction of key subnetworks: this can be used for uncovering routes to pathogenesis, patient stratification, designing treatments, and as a platform for detailed models, which it can be used in conjunction with. It is well-established that misfolded alpha-synuclein (alpha-syn) is a histopathological marker of Parkinson's disease. Given this, we focus on key pathways centered around alpha-syn to explore the pathogenesis of Parkinson's disease. These are (i) the reactive oxygen species (ROS) pathway, which involves positive feedback between misfolded alpha-syn and oxidative stress, and (ii) the proteasomal feedback pathway involving a negative feedback between misfolded alpha-syn and the protein degradation machinery (also see Supplementary Material: Appendix S1). A study of the individual pathways (see Figure 1) reveals that (a) the ROS pathway can result in both sharp switching and graded responses, depending on parameters such as the basal ROS production, alpha-syn misfolding rate, strength of positive feedback, etc.5 and (b) the proteasomal feedback pathway can result in both steady-state output as well as different kinds of oscillations (see below), depending on the strength of the negative feedback as well as binding and catalytic rate constants involving interaction of proteasomal components and alpha-syn aggregates.6 The combined effect of the positive feedback in ROS pathway and negative feedback in the proteasomal pathway can lead to qualitatively different outcomes. The positive feedback in the ROS pathway can contribute to elevated levels of misfolded alpha-syn corresponding to a diseased state. This could happen either gradually or via a bistable switch, associated with a region of irreversible switching. By contrast, the proteasomal feedback pathway can result in either steady-state behavior or sustained oscillations in misfolded alpha-syn levels which are also reflected in the levels of proteasome sequestration. Depending on the mechanism generating oscillations (supercritical or subcritical Hopf bifurcation), this could be characterized by phases of high misfolded alpha-syn and high levels of sequestered proteasome and irreversibility (subcritical case) or reversibility (supercritical case). Our analysis reveals that the emergent behavior arising from the contributions of these pathways can involve any combination of A1 (gradually increasing alpha-syn level), A2 (bistable switch in alpha-syn level), B1 (no oscillations), B2 (oscillations arising from a supercritical Hopf bifurcation), B3 (oscillations arising from a subcritical Hopf bifurcation). Furthermore, in the instances involving a combination of a bistable switch (A2) and oscillations (either B2 or B3), our analysis reveals two essentially different scenarios: one where the regions of bistability and oscillations overlap in parameter space and one where they do not. This, in totality, reflects the essentially different emergent outcomes from the interaction of these subnetworks which could correspond to different patient subpopulations (see Table S1 for a characterization of different kinds of emergent outcomes). Here, we particularly focus on pathology arising from a bistable switch in alpha-syn levels (A2) and interrogate system response to both step and pulse inputs (changes to key system parameters; see Table S1). At the outset, we note that in the absence of oscillations (B1) both step and pulse inputs are capable of triggering pathology (elevated alpha-syn levels; Figure 2f,g). If oscillations are possible, then we have two different scenarios (see Figure 2a–e): (1) a pathological state characterized by both elevated alpha-syn levels and proteasomal sequestration can be triggered by both step and pulse inputs (corresponding to overlap between bistable and oscillatory parameter regions), and (2) where this can be triggered only by step inputs (see Figure S1A–D: this corresponds to no overlap between bistable and oscillatory regions). In the latter case, pulse inputs can still trigger a pathological state with elevated alpha-syn levels alone. If oscillations are triggered by a subcritical Hopf bifurcation, then a pulse input is capable of triggering these oscillations whether or not the alpha-syn dynamics are characterized by a bistable switch (as long as the oscillatory region extends to the basal parameter state, co-existing with the steady-state) indicative of an additional layer of irreversibility in the system, evidenced by a pulse triggering oscillations (Figure S1E). We note that the case of a graded response leading to elevated alpha-syn levels is discussed in Table S1. In this case, the potential for irreversibility (and associated switching by a pulse input) is entirely dependent on the presence of a subcritical Hopf bifurcation. All this analysis reveals the significantly different routes to pathogenesis, their associated characteristics, and what types of inputs may trigger them. These could, in theory, correspond to different patient subpopulations. We note that depending on the patient subpopulations (as determined by network parameters), different types of intervention and therapeutic targets are possible (also see Supplementary Material, Section 2). We explore this briefly below to show how a pathological state involving elevated alpha-syn levels (associated with a bistable switch) and oscillations (of either type) can be reversed. We build on the dynamical characterization above to examine two potential therapeutic targets which are basal ROS production and proteasome availability. The impact of a step change in ROS production rate can be directly inferred from the bifurcation diagrams in the above analysis. Owing to the irreversible nature of the bistable switch, lowering ROS to basal levels (normal physiological levels) may not be sufficient to reverse the switch to a low alpha-syn state, but may be capable of eliminating oscillations/proteasome sequestration. Increasing proteasome availability can have the following impact: an increased threshold for the bistable switch creating an additional buffer against switching to the diseased state; lower alpha-syn levels in the diseased state; it can also bring a degree of reversibility to the bistable switch to the diseased state. This allows us to directly demonstrate the combined impact of these parameters in reversing the diseased state: a step change in proteasomal availability along with a pulse reduction in ROS production. As seen in Figure S2, this combined intervention is able to completely reverse the pathological state. Crucial to this is the effect of proteasomal availability introducing a degree of reversibility in the bistable switch (see Figure 2i), which then allows a pulse reduction in ROS production to facilitate an exit from the diseased state. This demonstrates how a careful consideration of the dynamical characteristics of the subnetworks and their interactions inform potentially efficacious and unintuitive treatment possibilities. Our bottom-up approach (highlighting our systems perspective of unraveling crucial aspects of network behavior from interaction of key subnetworks) reveals the landscape of emergent systems behavior and alternative/fundamentally different routes to pathology (Table S1). We start with core components and subnetworks and use this to demonstrate a fundamental set of transitions underlying pathogenesis (also see Supplementary Material 2: Comments on Models). Because these are central components, they directly reveal important behavioral aspects of the network. We do not focus on a comprehensive model and, in fact, the existing framework serves as a platform for incorporating further details and calibration. These details could either have a modulatory quantitative effect (along the lines observed) or even a modulatory qualitative effect (further underscoring the importance of using the above as a platform). Nonlinearity is the norm in cellular networks and disease pathogenesis and progression are strongly nonlinear phenomena. Furthermore, stratification of disease progression and outcomes in patients is already an approach used in the field.7 Thus, a bottom-up approach for dissecting the behavior of these nonlinear networks starting from constituent subnetworks and using this as a basis for both organizing analysis of network behavior and stratification of patients could be very valuable in systems medicine/pharmacology. This can benefit from prior foundational work.8 In light of our analysis, the core subnetworks constitute a key skeleton of the network, and a particularly appropriate platform for building detailed models. Whereas in Parkinson's disease, the subnetworks share a key component (alpha-syn), in other instances, key subnetworks could interact with or without shared components, and could be analyzed similarly, describing the interaction in a coarse-grained way. There are different physiological sources to Parkinson's disease recognized in the literature. Along another axis, we reveal different qualitative routes to pathogenesis as well. Going forward, a systems pharmacology approach needs to coherently integrate these two axes to sharply connect different physiological sources and their qualitative manifestations. Our analysis reveals potentially different routes to pathogenesis (Table S1) and different patients could be on different regions of the systems landscape. By building on the modeling platform here, on one hand, and improved and focused data collection and design of experiments on the other, there is potential for using this for a mechanistic systems-based patient stratification in the future. Our analysis has many implications for treatment. It reveals multiple sources of partial irreversibilities in the network, which play a critical role in shaping treatment. An examination of the network indicates which nodes to target: in particular, we reveal via an analysis that by targeting two nodes together, we can have an efficacious treatment, even in the presence of irreversibilities (and associated emergent behavior) and this softens the requirement on one dose/target. By focusing on the key underlying structures, we obtain transparent insights into how a treatment may be shaped (also see Supplementary Material). This can be used as a platform for a detailed calibrated model which can be the basis for quantitative analysis and define an optimal protocol for treatment. Importantly, alongside a mechanistic approach to patient stratification, this can be used to tailor treatments in a patient-specific manner. Our analysis of dopamine compartmentalization builds on this platform, describing a factor implicit in the earlier modeling framework in more detail (Supplementary Material). This reveals a clear structural feature: built-in competing effects and trade-offs associated with dopamine. The balance of these competing effects could potentially vary in a patient-specific manner, and suggests there could be patients in whom the addition of dopamine may be unfavorable. In clinical practice, although dopamine treatment typically improves patient symptoms, some clinical data show that it may worsen neurodegeneration.9 Overall, our approach is a systematic basis for dissecting systems complexity, revealing the systems landscape of behavior, mechanism-based stratification of patients, investigating systems medicine treatment options, and creating a platform and tool/methodology for further analysis/investigations. No funding was received for this work. The authors declared no competing interests for this work. Appendix S1 Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.